Digital-twin-based Auto-leveling Control and Anomaly Detection Methods for Drawing Frame
Supervisor:chen you ping chen bing
The development of intelligent textile equipment is the key to improve the modern textile industry.The research in this thesis aims to improve the performances of autoleveling control and product quality monitoring using industrial intelligent computing abilities of Digital Twin techniques.In this way,the quality of textile product can be elevated in a significant margin.This thesis focuses on the research topics including speed control,control-parameter prediction and abnormal detection methods.According to the research,following problems and corresponding approaches are summarized as the chock points of the mentioned methods and the scientific contributions proposed in this thesis.Firstly,a Digital Twin virtual sensor-based observer is used to solve the load torque unobservable problem in speed control method.The precision and robustness of speed control are improved based on the proposed method.Secondly,a Digital Twin intelligent computing based method is applied to solve the environment and control parameters modelling and computing time problem in control parameter predicting method.The proposed control parameter predicting method is adaptive to various environments and runs in real time.Thirdly,a Digital Twin intelligent monitoring-based method is used to deal with the abnormal channel detecting and precision problems of the abnormal detection.The proposed method improves the computational speed and accuracy of the abnormal detection in a significant margin.To verify the methods proposed in the thesis,an intelligent control and abnormal detection testbed is built using the edge computing of Digital Twin framework.The contributions of the research in this thesis include:(1)A robust speed control method using the Digital Twin virtual sensor techniqueDuring the auto-leveling control in drawing frame,the resonance is caused by the unexpected resonant frequency in the shaft with compliant joints.Meanwhile,the precision and robustness reduce in speed control due to the highly non-linear load torque.The difficulty of speed control is that the load torque is highly non-linear and unobservable,which results in a poor performance for the speed control method in the shaft with compliant joints.To solve the problem,a non-linear extended Kalman Filter is designed to obtain the load torque using the virtual-sensor method in Digital Twin framework.The load torque can be estimated via measurable output speed with the proposed virtual-sensor.Combining with the LQG controller,the speed control error caused by varying load can be reduced.(2)A control parameters predicting method using the Digital Twin intelligent computing techniqueDuring the auto-leveling control,the thicknesses of the input materials change rapidly and the tension of the materials also varies frequently.These properties result in a low precision and poor robustness of the control parameter prediction method.The difficulties of control parameter prediction are that: Firstly,the relationship between control parameters and the environmental information is hard to be established.Secondly,the method has to be computational efficiency so the rapid changing information is traceable.Inspired by the intelligent computation of Digital Twin technique,a wavelet neural network is applied to establish the relationship between the environment and control parameters in the data-driven fashion.In this way,the control parameters are self-adjustable.Combined with the nonlinear general predicting control,the proposed method yields a higher precision and better robustness.(3)A abnormal detection algorithm based on the Digital Twin intelligent monitoring techniqueIn the online detection system of drawing frame,the quality monitoring method needs to detect the abnormal data points from a large number of potential abnormal channels.Thus,the abnormal detection method results in a low accuracy.The difficulty of detecting abnormal points from the spectrogram is that the number of normal channels is much greater than the number of abnormal channels.In this way,the signal noise ratio(SRN)of the abnormal data point is low and the algorithm has trouble to capture the abnormal information.Inspired by the intelligent monitoring of the Digital Twin technique,a random sample consistency based abnormal detection method is proposed.The SRN of the abnormal data point is elevated using the detecting consistency property of the unsupervised isolated forest.(4)A testbed for auto-leveling control and product quality monitoring system for drawing frame based on the Digital Twin techniquesA testbed for auto-leveling control and product quality monitoring system for drawing frame is built with Digital Twin hardware and software.The proposed speed control,control parameter predicting and spectrogram abnormal detection methods are tested and verified using the testbed.The experiment results show that the proposed methods in this thesis can improve the quality of the sliver and have a high product quality abnormal detecting accuracy.The textile manufacturing process is digitalized and modernized with the proposed methods.